Capital Scenarios Methodology
A bitcoin treasury company funds its BTC with common stock, preferred stock and convertible notes. When BTC moves, the effect on the common stock is not a fixed multiple of the BTC move: the premium to BTC holdings (mNAV) moves too, the company issues or stops issuing shares, pays dividends and interest from a USD reserve, may sell BTC, and holders of convertible notes decide between conversion and a cash put. The capital scenario grid answers one question for each combination of a BTC price move and a horizon: what happens to the stock, the capital structure and the company's BTC buying or selling if BTC gets there?
The grid is an Enterprise feature. The pilot company is Strategy (ticker MSTR). It is recomputed once a day.
Scenarios are model outputs based on public filings, market data and stated assumptions. They are not forecasts of future prices, not investment, legal or tax advice, and not an offer or solicitation to buy or sell any security. Company policies may change at any time. Past behaviour does not guarantee future results.
1. The grid
| Axis | Values |
|---|---|
| BTC move from today's close to the horizon | −60%, −50%, −40%, −30%, −20%, −10%, 0, +10%, +25%, +50%, +100% |
| Horizon | 1, 3, 6 and 12 months (4, 13, 26 and 52 weeks) |
Each of the 44 cells is simulated with 4,000 BTC paths. Distribution outputs are reported as the 10th, 50th and 90th percentiles across paths (p10 / p50 / p90) and the mean; events are reported as the share of paths in which the event happens at least once within the horizon.
2. BTC paths
A scenario fixes where BTC ends, not how it gets there. The route matters: a −40% horizon reached through −60% can stop share issuance or trigger BTC sales that a straight decline would not. Each path is a random walk of weekly log returns pinned to the target price (a bridge):
where is the BTC move, the horizon in weeks and Student-t shocks with 4 degrees of freedom rescaled to unit variance (BTC has jumpy weeks). The weekly volatility comes from the current regime: = ½ × 30-day + ½ × 365-day realised BTC volatility, kept between 25% and 120% a year. The same random shocks are reused for every move of a horizon, so neighbouring cells differ only by the move.
3. mNAV model
(basic). Week by week:
- — the BTC weekly log return of the path. is the immediate sensitivity of the premium to BTC: with the stock moves about 1.44 times the BTC move in the short run.
- pulls mNAV back towards its long-run level with a half-life of about 23 weeks.
- — historical weekly residuals, resampled in blocks of 8 weeks (keeps fat tails and clustering).
- — a slowly drifting level ( a week): the premium regime itself can shift.
Parameters are estimated by least squares on weekly closes from January 2022 (earlier, the software business dominated the market value relative to the BTC holdings). Stock prices are split-adjusted and the share count is on the same basis.
How it was chosen. Several specifications were compared out of sample on 2023–2026 with an expanding window (each forecast uses only data available at its date, and is conditioned on the BTC path that actually happened): a constant mNAV, a naive beta to BTC, the model above, and richer versions with momentum, volatility, drawdown, issuance pace, asymmetric response, a moving anchor or a memory of past BTC returns. The richer versions fitted the past better and forecast worse. 2022 was used as a stress test with parameters fitted on 2024–2026 only.
| Horizon | 1 m | 3 m | 6 m | 12 m |
|---|---|---|---|---|
| mNAV error, model (2023–2026) | 0.15 | 0.27 | 0.39 | 0.55 |
| mNAV error, constant mNAV (2023–2026) | 0.16 | 0.24 | 0.36 | 0.61 |
| Stock error, model (2022 stress) | 0.18 | 0.26 | 0.28 | 0.15 |
| Stock error, naive beta (2022 stress) | 0.21 | 0.29 | 0.22 | 0.20 |
| p10–p90 band coverage, model (2023–2026, target 0.80) | 0.90 | 0.83 | 0.82 | 0.75 |
Errors are root mean squared errors of log values (0.15 ≈ 15%). The model's main advantage is robustness: it is never far off in the 2022 crash and beats a constant premium at 12 months. In the 2023–2024 rally a naive beta had a smaller stock-price error at 6–12 months, because the premium kept expanding with BTC; the grid shows the naive-beta price next to the model for that reason.
4. Capital simulation
On every path the company's balance sheet is walked week by week, starting from the latest reviewed filings: BTC holdings, basic shares, the USD Reserve and other USD cash, every convertible note (principal, coupon, conversion price, holder put and maturity dates) and every series of preferred stock (stated amount, dividend rate, seniority). In each week:
- Price. Stock price , with mNAV from §3.
- Obligations. Preferred dividends and note interest accrue weekly and are paid from the USD Reserve, then from other USD cash. When mNAV is below 0.75 they are paid by selling BTC instead (the company did this in 2026).
- Convertible notes. On a holder put date (or maturity) holders convert into shares if the stock is above the conversion price; otherwise they put the notes for cash. A put is paid from USD cash, then from a share sale of at most 5% of market value (only if mNAV ≥ 0.75), then by selling BTC. Refinancing is not assumed. A claim that cannot be paid even after selling all BTC stays in the capital stack as an unpaid claim (default).
- Reserve minimum. The USD Reserve must cover at least 12 months of dividends and interest (company policy). A shortfall is refilled from share sales, or by selling BTC when share sales are off.
- Share sales (ATM). Weekly common stock sales are a share of weekly trading volume that rises with mNAV: none below 0.75, 2% from 0.75, 5% from 1.0, 8% from 1.5 and 12% from 2.5. Above mNAV 1 the proceeds buy BTC; below 1 they are split between BTC (30%), preferred repurchases (65%) and USD cash (5%), as in the company's recent reports.
- Preferred stock. When credit conditions are calm (BTC less than 15% below its 13-week high — counting the weeks before today — and mNAV ≥ 0.80) the company issues preferred stock at par (about $190M a week, its 2026 average) and buys BTC. Otherwise it repurchases preferred stock below par within the remaining authorisation.
- Feedback. The mNAV after the week's trades (share sales, BTC bought or sold) is the starting point of the next week.
Rules 2–6 paraphrase the company's published capital framework (USD Reserve policy, dividend rate policy, preferred and common repurchase programs, BTC monetization program) and its weekly reports; the thresholds that the company does not state are calibrated on its 2024–2026 behaviour. They live in an editable policy file with a source for each rule and are reviewed after every quarterly report.
Forced vs policy BTC sales. A forced sale is one needed to pay obligations, a note put or the reserve minimum
when no other source is available. A policy sale is a sale the company chooses under its program (rule 2, or
funding repurchases when mNAV is below 0.75). Both count in BTC sold; only forced sales count in
prob_forced_btc_sale.
5. Outputs
| Output | Meaning |
|---|---|
stock_price, stock_price_naive_beta | Stock price at the horizon: model, and today's price × (1 + move)^β with the trailing 52-week beta (comparison) |
mnav_basic, mnav_ev | mNAV basic, and enterprise-value mNAV: (market cap + debt + preferred − USD assets) / BTC value |
btc_per_share, btc_holdings, shares_basic, dilution_pct | BTC per basic share, holdings, shares and share growth since today |
usd_reserve, runway_months | USD Reserve; months of dividends and interest covered by USD assets |
coverage_<layer> | (BTC value + USD assets) / all claims senior to and including the layer: notes, then each preferred series by seniority |
btc_bought, btc_sold, btc_sold_forced, btc_sold_usd | BTC flows over the horizon |
btc_sold_pct_spot_volume, btc_sold_peak_week_pct_spot_volume | BTC sales as a share of BTC spot trading volume over the horizon, and the largest week |
mstr_atm_usd, mstr_atm_pct_adv | Common stock sold and its share of the stock's trading volume |
pref_issued_usd, pref_repurchased_usd, convert_put_usd | Preferred issued, preferred repurchased, notes put for cash |
prob_mnav_below_1, prob_mnav_below_1_end | mNAV below 1 in any week / at the horizon |
prob_btc_sale, prob_forced_btc_sale | Any BTC sale / a forced BTC sale |
prob_convert_put | A note put for cash (stock below the conversion price on a put date) |
prob_reserve_below_min | USD Reserve below its policy minimum in any week |
prob_default | Cash claims left unpaid after all BTC and USD assets are used (the stock is then valued at 0) |
Coverage, runway and enterprise-value mNAV can be unbounded on some paths (a layer fully repaid, no obligations left, no BTC left). For these outputs the probability field gives the share of such paths, and a percentile that falls in that part is left empty instead of being dropped from the distribution.
6. Data
- Capital stack: the company's SEC filings (10-Q, 10-K, 8-K, prospectus supplements), parsed daily and reviewed by a person before use; only approved records enter the model. Balances from the last quarterly report are rolled forward with the weekly reports (share sales, repurchases) and the latest declared dividend rates.
- Missing terms: if a small instrument has no rate, amount or seniority in the filings (for example a floating-rate loan), the model fills it from a documented assumption — the rate formula in the filing with an assumed reference rate, the remainder of the reported debt total, secured loans ranked first — and lists it with the grid's inputs. If such instruments are more than 1% of debt and preferred stock, no grid is published for that day.
- Market data: daily closes and volumes of the common stock, the BTC price, BTC spot trading volume and the company's BTC holdings.
- Calibration: weekly mNAV history from January 2022. Parameters are re-estimated by hand after quarterly reports, not daily; the daily run uses the saved parameters and today's balance sheet.
7. API and MCP
GET /v1/dat/capital/scenarios?entity=MSTR (Enterprise) returns the latest grid: the inputs snapshot, the model
block, a dictionary of metrics with units, one row per cell × metric (p10, p50, p90, mean, prob), the
available dates and the disclaimer. Filters: date, horizon_m, metrics. The MCP tool
get_dat_capital_scenarios wraps it with a compact per-cell view.
8. Limitations
- Policies change. The rules describe what the company says and has done; it can change them at any time (it did several times in 2025–2026). Thresholds it does not publish are estimates.
- The premium is the hardest part. mNAV errors grow with the horizon (about 55% at 12 months). The p10–p90 bands are wide on purpose and still covered the outcome only about three times out of four at 12 months in 2023–2026.
- Mean reversion is an assumption. With BTC flat, the model lets mNAV drift back towards its 2022–2026 level; if the current regime persists, that is optimistic.
- Simplifications. Preferred prices, the dividend rate of the variable-rate preferred (held at the current rate), refinancing of notes, operating cash flow of the software business, taxes and fees are not modelled. Issuance of new kinds of securities is not assumed.
- Market impact is not modelled. BTC sales and share sales are reported as a share of volume; their effect on prices is left to the reader.